Bruce Carsten Assoc. · 6410 NW Sisters Place · Corvallis, OR 97330 · (541)745-3935
This article is reproduced byPOWERTECH, INC. with the permission of Intertec International Inc.
It was presented in the proceedings of the PCIM conference in 1993.
There is an ongoing effort to develop improved solid state switches as converter power levels and conversion frequencies rise. The Bipolar Junction Transistor (BJT) was the first commercial solid state switch, followed by the SCR, GTO, FET and others. As the first, BJTs are often considered a mature and even aging technology, with superior performance obtainable from the newer devices. On the contrary, the ability to control conductivity modulation gives bipolar transistors a potential speed/power advantage over other silicon semiconductor switches in certain applications. In principle today's technology could achieve blocking voltages of 1-2 KV Vcbo, 200 to 500 Amps conduction currents and switching speeds of 10 to 100 ns. Construction and operation of the various power switches are compared to illustrate the advantages of the BJT, with proposed approaches to the realization of the BJT's potential.
Power electronics has been growing rapidly for the past twenty years or so, expanding into areas ranging from palmtop computer power supplies to megawatt motor drives. This expansion into new markets is governed by evolving needs, economics and theoretical and technical limitations.
Solid state switches are presently a major constraint on the high power/high frequency/high efficiency corner of the power conversion performance envelope.
The "Missing Link"
There is a growing requirement for 1-2 KV semiconductor switches with conduction current ratings of 100 to 500 Amps at a few volts forward drop. More importantly, switching must be fast and controllable, with turn-on and turnoff times of less than 100 ns achieved over a range of operating conditions.
Perhaps the most immediate applications are in the induction heating field. Frequencies of 100 to 500 KHz and power levels of 30 KW to several Megawatts presently require the use of large, expensive and inefficient vacuum tubes. The proposed solid state switch would open up this market fairly quickly. Other near term applications would be in sine-wave-synthesizing DC-AC inverters for Uninterruptable Power Systems and light weight traction motor drives.
Longer term applications would be in light weight cycloinverters to generate constant voltage/constant frequency power from jet engine and other gas turbine alternators (eliminating the variable ratio transmission), high frequency arc welding and plasma generation, AC mains harmonic current reduction and power factor correction, and industrial laser drivers. Eventual applications could range from power grid stabilizers to ion rocket drives.
Characteristics and Limitations of Existing Switches
Historically the emphasis has been on the ease of switch driving (which simplifies circuit design) and on high current density (for economical use of silicon), with switching speed a secondary concern. The first solid state power switch worthy of the name was the SCR, which is easily turned on and has a practical current density much greater than BJTS. Both desirable properties are a consequence of internal drive regeneration, which latches the SCR on. The principal drawback in DC applications is that the current must be externally brought to zero, or K-A voltage reversed, to turn the device off.
Their simpler turn-off kept bipolars in use in many lower power DC applications, and higher voltage and current ratings became available with time. The need for a significant steady state drive current was still considered inconvenient, and their relatively low current densities limited power handling capability. The development of Darlington transistors reduced drive current significantly and increased practical current densities somewhat, at the expense of a moderate increase in conduction voltage.
Attempts to overcome the limitations of BJTs and SCRs resulted in the development of other power switch technologies including FETS, GTOS, IGBTS, SITs and recently MCTS.
In order to understand the limitations of semiconductor switches, and the potential advantages of BJTs as high speed/high power switches, a superficial review of semiconductor physics and the operation of various devices is in order.
Device Physics Limitations
In the future the use of exotic semiconductor materials, such as silicon carbide or diamond, may well allow vast improvements in solid state switch performance. For at least the next ten years or so however, silicon will probably remain the only practical power semiconductor material.
A relatively thick, lightly doped (i.e., highly resistive) region in a semiconductor is required to sustain high "off state" voltages. Both the resistivity and minimum thickness of the lightly doped region must increase to prevent avalanche breakdown. The breakdown voltage gradient in silicon also decreases with reduced doping levels. The net result is that, with optimized doping and thickness of the high resistivity region, the minimum resistance per unit area varies as the 2.5 power of the breakdown voltage, as shown in Figure 1.
The resistance of the lightly doped region is thus quite high in high voltage devices. In majority carrier devices such as FETS, the minimum forward voltage drop during conduction is the product of this resistance and the conduction current, severely limiting current densities due to IIR heating.
BJTS, SCRS, IGBTs and other solid state switches overcome this limitation through conductivity modulation; minority carriers injected into the lightly doped region during conduction create an electron-hole plasma, greatly increasing the "effective doping" of charge carriers available for conduction, lowering the resistance by orders of magnitude. With lower conduction drops, current densities can be increased by 10 to 100 times or more.
However, a price is paid in reduced switching speed for this lower conduction resistance. Conductivity modulation is not instantaneous; a finite time is required to inject the required minority carrier charge and distribute it throughout the lightly doped region. Excess conduction loss occurs as conductivity modulation proceeds and the forward voltage lowers to the steady state value.
When the switch is to be turned off, additional time is required to reverse the process and remove the "stored" charge before the device can again sustain full off-state voltage. If the device is forced off with current flowing, the conduction voltage rises as the charge is removed, further increasing power loss. Often full voltage is reached before all charge is removed and current ceases to flow, which typically makes losses at turn-off greater than at turn-on. (This high voltage "current tailing" during turn-off can also create potentially destructive local stresses in the switch, which are beyond the scope of this paper.)
Furthermore, if the minority charge injected is greater than required to achieve a low conduction resistance there is little or no benefit in reduced conduction loss, but the excess stored charge will delay and often slow down turn-off.
If the benefit of conductivity modulation (low steady state conduction loss) is to be achieved at the lowest cost (fast switching speed and minimized switching losses), the following somewhat overlapping goals must be met:
1) Minimize the injected minority charge required to lower the resistivity;
2) Optimize the distribution uniformity of charge injection and removal (i.e., minimize the lateral diffusion distance);
3) Minimize the vertical diffusion distance;
4) Prevent the injection and storage of excess minority charge.
In other words, minimize the charge required, get the charge in and out as fast as possible, and avoid excess charge injection and storage.
The mechanism of minority carrier injection and extraction, and the degree of control during switching, differs between switch technologies. Various semiconductor switches will be reviewed to see how well they meet the criteria.
Field Effect Transistors
Although Field Effect Transistors do not normally exhibit conductivity modulation, they can form part of switches which do, and are reviewed here for completeness.
MOSFETs have an insulated control gate and are usually normally-off devices (Figure 2). For an N-channel device, a positive voltage on the gate creates an N-channel in the lightly doped P region under the gate electrode, which creates an ohmic connection between the N doped source and drain. In high voltage devices reductions in channel resistance at higher gate voltages are overshadowed by the resistance of the lightly doped drain region, but the "saturation current" at high drain voltages remains dependent on gate voltage, as illustrated in the characteristic curves of Figure 3.
Power JFETs (Fig. 4) are normally-on devices with a junction isolated gate. A negative gate voltage creates a depletion region around the gate which "pinches off" the conduction channel between source and drain. The control characteristic (Figure 5) is quite different from MOSFET'S. For a given negative gate bias drain current is prevented until a drain voltage is reached where the current increases exponentially.
A Static Induction Transistor (SIT) is basically a power JFET with a buried gate, as shown in Figure 4b. Construction as well as operation is reminiscent of a vacuum tube.
As a class, FETs naturally switch very quickly due to the lack of conductivity modulation.
Bipolar Junction Transistors
The construction of bipolar transistors shown in Figure 6 is familiar to many. A typical high voltage NPN transistor has a heavily N+ doped emitter, a moderately P doped base region and a lightly N doped collector, with a Heavily N+ doped collector ohmic contact.
With a high voltage on the collector, the collector-base junction is reverse biased, and a charge depletion region extends largely into the collector due to its lower doping level. If a forward current is made to flow in the base-emitter junction, most of the current consists of electrons flowing from emitter to base, with a small current component of holes flowing from base to emitter, due to the higher N+ doping of the emitter relative to the P doping of the base.
With a thin base, most of the electrons from the emitter flow through the base region and into the collector, causing a collector current to flow which is typically much larger than the base current. As long as the collector voltage is sufficiently high the transistor operates in the "unsaturated" region of the characteristic curves of Figure 7, where the collector/base current ratio is approximately constant. Only electrons are injected into the collector from the emitter. As the collector is N doped these electrons are "majority" carriers, and collector current can change very rapidly in response to base current.
As the collector voltage lowers with the depletion region disappears and current flowing, a point is reached where the collector-base junction becomes for-ward biased (quasi-saturation begins). At high currents there will still be significant voltage (perhaps tens of volts) on the external collector contact due to the resistive IxR drop in the lightly doped region. The forward biased collector-base junction now begins to inject holes (minority carriers) into the lightly doped collector.
The presence of holes in the collector increases the electron density to maintain approximate charge neutrality, and the resulting electron-hole plasma lowers the collector resistance, and hence the voltage drop. If the base current is increased further the collector resistance drops roughly as the inverse of the base current, until eventually a point is reached where fixed resistances become dominant. Still higher base currents increase the stored charge in the collector region, but the collector-emitter voltage falls no further and the transistor is in full or deep saturation.
If the operating characteristics of similarly constructed MOSFETs (Fig. 3) and BJTs were superimposed, it would be seen that the "on" resistance of MOSFETs ("Ron" in Fig. 3 & 7) corresponds to the onset of quasi-saturation in BJTS. Conductivity modulation in the BJT allows operation to the left of the "Ron" line, with much lower conduction voltages.
Turning off the bipolar transistor requires removal of all the charge stored in the collector. Three mechanisms can remove charge from the bipolar's collector region:
1) Electron-hole charge recombination (finite charge lifetime);
Charge lifetime is normally on the order of 10 us or so, and is only significant when turn-off is very slow. Lifetime killers can be used, but this slows down conductivity modulation during turn on, increases steady state conduction loss, increases leakage current and lowers breakdown voltage.
2) Internal base-emitter current (charge supplied by the collector to maintain conduction);
Removal of forward base current from a saturated BJT stops the injection of minority charge into the collector, and the stored collector charge now flows into the emitter through the base, sustaining conduction. This internal base current is amplified by the current gain of the transistor, and is also a slow mechanism for turn-off.
3) External reverse base current;
Large reverse base currents are the most effective way to turn off BJTs quickly, particularly in conjunction with stored charge minimization schemes (such as avoiding deep saturation). There are various limitation on the amount of reverse base current that can be drawn, such as lateral base resistance, base-emitter junction breakdown voltage, and charge drift velocity from the collector into the base.
Silicon Controlled Rectifiers
SCRs are four layer/three junction devices as shown in Figure 8a. Their operation is typically illustrated by separating the structure into two interconnected complimentary BJTs as shown in Figure 8b, with the equivalent circuit shown in Figure 8c.
The two bipolar transistors are connected such that the collector of each drives the base of the other. A small current supplied to the base of the NPN transistor (the gate of the SCR) becomes amplified by the gain of the NPN and drives the base of the PNP, where it is further amplified and added to the original NPN base current. Thus a small initial gate current can quickly build to a very large anode current, which must typically be limited by the external circuit to avoid device destruction.
This regenerative drive process is so effective that steps are taken to prevent SCR turn-on every time the gate smells an electron coming. Transistor current gains are intentionally limited, particularly at low current, and base-emitter shunting resistances are built in, at least on the gated transistor (Rs in Fig. 8c).
The regenerative drive also maximizes conductivity modulation, so high currents can be conducted with low loss. Unfortunately, the same process makes SCRs and other thyristor devices difficult and slow to turn off.
Turn-off of an SCR requires that the current at least be brought to zero long enough for the stored charge to dissipate. When charge recombination effects dominate, several tens of microseconds are required unless lifetime killing techniques are applied. More commonly anode-cathode voltage is forced to reverse, which causes a high reverse anode-cathode current until charge carriers are swept out. This process is similar to reverse recovery currents in rectifiers, and results in a faster turn-off. In some devices a negative gate current can accelerate turn-off by a factor of two or three.
Gate Turn Off thyristors (GTOS) are essentially similar to SCRS, but are physically and electrically designed to allow sufficient negative gate current to be drawn to stop the conduction process, forcing device turn-off. This typically involves a finer gate structure interdigitated with the cathode, and a high current gain NPN transistor with low PNP gain, which allows reverse gate currents of 10-20% of the anode current to interrupt operation.
There are several additional thyristor structures in use which overcome one or more of the drive difficulties or limitations of the SCR and GTO, and these will be touched on lightly.
The MOS gated thyristor (Figure 9) uses an N-channel MOSFET in parallel with the NPN transistor to provide fast turn on with a high impedance drive. A positive gate voltage turns on the FET which drives the base of the PNP transistor, which in turn drives the NPN base and regeneration commences. Turn-off must still be achieved externally as with the SCR.
The MOS turn-off thyristor (not shown) uses a MOSFET structure to short out the base-emitter of the NPN transistor, which forces turn-off without the need to bring the anode current to zero or to reverse the anode voltage.
Both turn-on and turn-off are MOSFET driven in the MOS Controlled Thyristor (MCT). One possible structure is shown in Figure 10, with the equivalent circuit. An N-channel FET in parallel with the NPN transistor initiates turn-on, while a P-channel FET shorts the NPN base-emitter for turn-off. Thus a positive gate-cathode pulse turns the MCT on, and a negative gate pulse turns it off. Tile gate voltage may be zero during conduction and non-conduction. Complimentary devices and modified gating schemes are also possible.
Although these later thyristor types simplify turn-on and/or turn-off and achieve high current densities, conductivity modulation is a regenerative process in all of them, and is not under circuit control.
However, there are other switch structures besides the BJT which accomplish conductivity modulation without regeneration.
Insulated Gate Bipolar Transistors
The structure of these increasingly popular devices, shown in Figure 11, is similar to an N-channel MOSFET with the N+ doped drain region changed to P+. When the MOSFET is turned on by a positive gate voltage, the electrons entering the P+ region cause a heavy hole current to flow back into the lightly doped N region, modulation conductivity. The IGBT thus behaves like an N-channel FET driving a PNP bipolar transistor, as shown in the equivalent circuit.
The IGBT is also very similar to the MOS gated thyristor of Figure 9, except drive regeneration at normal current densities is prevented by a very low base-emitter shorting resistance on the MOSFET's parasitic NPN transistor.
The device can be turned on and off as a MOSFET, with the lower conduction loss of a MOS driven bipolar. Turn-off is somewhat slower than possible with a BJT, since there is no negative base current in the PNP transistor; internal recombination and effective collector-base current are the only available means to removing the stored charge. Lifetime killers may be used to achieve faster turn off times, with a resultant higher forward drop during conduction.
The MOSFET does prevent the PNP collector-emitter voltage from falling below the base-emitter voltage, which prevents deep saturation of the PNP.
Field Controlled Diodes
Two similar structures known as Field Controlled Diodes (FCDS) are shown in Figure 12 with their equivalent circuit. The construction is similar to that of Power JFETs (Figure 4) with the N+ drain region changed to P+. This is the same transformation that converts a MOSFET into an IGOT, and the equivalent circuit is indeed that of a JFET driving a PNP bipolar, and conductivity modulation is achieved with JFET drive characteristics.
Conductivity modulation control is perhaps a little better with the FCD than with the IGBT, but control is still indirect; the gate voltage controls the JFET current which is in turn the PNP transistor's base current. Production variations in control characteristics may be a problem, but the normally-on property is perhaps the greatest stumbling block to most design engineers.
Bipolar Mode JFETS
It is possible to operate Power JFETs (see Figure 4) with a forward biased gate. This causes current to flow in the gate-source diode, which injects minority carriers into the drain region and achieves conductivity modulation. (Significant gate currents require the low resistance surface gate structure of Figure 4a.) The degree of conductivity modulation is directly related to the gate current, and it would appear that such a device could compete with BJTs for controllability of conductivity modulation.
This is perhaps true in a sense, but I see no particular advantage over BJTs. Indeed, the buried gate JFET of Figure 4b looks to me like a BJT with a perforated base which allows the collector to contact the emitter. This requires the usual negative gate JFET bias to prevent conduction, which the BJT does not, while the characteristics with forward gate drive (= base drive) currents should be similar for devices with comparable constructions.
The purpose of the above survey was to determine which device technology(ies) showed the greatest promise for low loss, high speed power switches. At the high voltages under consideration (1 KV and higher), conductivity modulation is necessary to achieve low loss and acceptable current densities during conduction. Achieving fast switching then requires the greatest degree of control over conductivity modulation under various operating conditions.
Although MOSFETs and Power JFETs are very fast switches, they do not exhibit conductivity modulation and are not suitable for the power and voltage levels considered here.
Conductivity modulation in SCRs and the other regenerative thyristor structures is completely uncontrolled, and generally leads to much more injected and stored charge than required to achieve low conduction loss. Switching times are relatively long and cannot be modified by circuit control.
Conductivity modulation in IGBTs and FCOs can be controlled in principle, but modulation is indirect, and a very strong and poorly defined function of the control voltage. Fine control of conductivity modulation is thus difficult in practice, rather like trying to control the collector current in a BJT by forcing the base-emitter voltage.
Conductivity modulation can be controlled in the bipolar mode Power JFET, but only to the same degree as in the BJT. JFETs are also normally-on devices, and the gate must be reversed biased to maintain an off state. Any power switch which is "default on" presents practical design problems. This reverse bias can be difficult to achieve when power is first applied, for example.
Thus I propose that the relatively ancient Bipolar Junction Transistor is the most promising semiconductor switch for the high frequency/high power applications envisioned. Unfortunately, BJTs (and particularly high power BJTS) are typically constructed with 10-30 year old technology, and do not live up to their potential as fast, efficient high power switches.
It is true that even optimized high voltage power BJTs will require substantial and highly controllable base drive currents to realize their performance potential. Even with base currents a significant fraction of the collector current however, the drive power can be less than 1% of the output power. Design expertise and sophistication will definitely be required, but will allow power converter performance that could not otherwise be achieved. All we really lack are the parts.
Although I am not a solid state physicist, I have a few suggestions as to how this potential might be realized.
Existing high voltage bipolar power transistors suffer principally from two "grandfather " fabrication techniques which limit performance; the collector regions are of diffused construction, and emitter geometries are extremely coarse. (Both of these are sometimes justified as producing rugged transistors, but "ruggedness" is irrelevant in sophisticated designs than can avoid transistor overstress under all operating conditions.)
Diffused collectors are much thicker than needed for the breakdown voltage achieved. A greater injected minority charge is needed to lower the (higher) resistance of the thicker collector, and it takes longer for the charge to diffuse in and out at low collector voltages. Thus excessively thick collector regions cause high dynamic conduction loses and slower switching.
Coarse emitter geometry limits utilization of the silicon to conduction along the emitter periphery, while allowing injected charge to diffuse under the emitters. This also slows the turn-off process in particular, and limits current gain and practical current densities.
Collector junctions should be epitaxially grown instead of diffused, allowing lightly doped regions of optimum resistivity and thickness for the required voltage. This process is currently used for FETS, RF bipolar transistors and the fastest rectifiers. Epitaxial collectors would minimize the time and injected charge required for conductivity modulation, speeding up switching and lowering dynamic conduction losses.
Fine emitter geometries (or the functional equivalent) would also improve dynamic gain and operating current density, and minimize turn-off losses due to current tailing from "hidden" charge deep under wide emitters.
Fine emitter geometries would make high current connections to the emitter, as well as low impedance connections to the base, far more difficult. A "mesa" emitter contact with recessed base metalization is an option that would allow an emitter pressure contact (Figure 13a). A two level base and emitter metalization might also work (Figure 13 b), either with a pressure emitter contact or a nest of bond wires.
A third alternative is to retain physically wide emitters, with heavily doped base current "distribution rails" beneath the emitters. A sketch of the proposal is shown in Figure 14.
The mask geometries required would still be quite coarse by modern FET and integrated circuit standards, and -the proposed bipolar transistors could probably be built with "second-hand" equipment.
Various semiconductor switch technologies were surveyed for their potential as fast high voltage/high current switches.
The switching speed of FETs can be controlled over orders of magnitude, but there is no conductivity modulation to lower on-state resistance. On the other hand, carrier injection in SCRS, MCTs and other thyristors is regenerative and not under circuit control; it can only be modified by changing the device design and construction.
Conductivity modulation in IGBTs and FCDs is not regenerative, but is only indirectly related to the gate voltage. Control of conductivity modulation during conduction would be difficult, and stored charge removal during turnoff cannot be accelerated with reverse gate drive.
Although the bipolar mode JFET allows good control of conductivity modulation, it's normally-on characteristic places it at a significant disadvantage.
It was found that the bipolar transistor has the best combination of characteristics, including direct control of minority carrier injection, ability to accelerate turn off with reverse base current, and normally off behavior.
Since minority carrier injection and removal can be fully controlled by the base current, bipolars have the potential to be the fastest conductivity modulated switches obtainable. I propose that their capabilities as high speed, high power switches has never been fully realized, and the task remains to achieve this potential through the application of available semiconductor technologies.
Uncountable papers and articles have been written on the various semiconductor switches, their construction, mode of operation, relative merits and so on. Instead of attempting a representative list, I direct the reader to the following three books which I found very informative and useful (and have extensive bibliographies of their own).
All of the following books are published by:
WILEY INTERSCIENCE, A division of: John Wiley & Sons, 605 Third Avenue, New York, NY, 10158-0012
 S. M. Sze, "PHYSICS OF SEMICONDUCTOR DEVICES", 2nd edition, © 1981
 Sorab K. Ghandhi, "SEMICONDUCTOR POWER DEVICES", © 1977
 B. Jayant Baliga, "MODERN POWER DEVICES", © 1987